Research On Image Super Resolution Reconstruction Algorithm Based On Deep Learning

Convolutional neural network learns end-to-end mapping between high/low resolution images,inputs low resolution images and outputs high resolution images.SRCNN with three convolution layers has a certain reconstruction effect,but its receptive field is low,and the reconstruction effect still has a lot of room for improvement.The residual network is introduced into the initial improved model RDSR-a,while the network structure is deepened and the size of convolution core is improved.On the basis of RDSR-a model,RDSR-b model is proposed.Local residual and deconvolution reconstruction are introduced to improve the effect of image superresolution reconstruction.The experimental results show that,compared with several existing reconstruction algorithms,the improved algorithm improves both objective and subjective evaluation indexes.Finally,the improved model is applied to the super-resolution reconstruction of compressed image,which achieves the effect of reconstructing high-resolution image.